1  

Remittances and the unwillingness to work in

Albania

Student: Dorela Beqaj

Student number: 372369

Supervisor: Dr. Sacha Kapoor

Date: 10 July 2017

Erasmus University Rotterdam

Erasmus School of Economics

Master Thesis International Economics

  2  

Abstract

As in many developing and emerging economies, remittances account for a considerable

share of GDP in Albania. This study investigates on the impact that remittances have in the

work incentives for remittance-receiver households, by using micro data for Albania in years

2009 and 2014. There results of Pooled OLS and Instrumental Variables indicate that

remittances positively affect the unwillingness to work for remittance-receiver households.

These findings are helpful for creating a clearer idea regarding the impact that remittances

have in individual’s (un)willingness to work in Albania.

  3  

Contents

1.Introduction…………………………………………………………………5

2. Literature Review…………………………………………………………...7

3. Background information…………………………………………………....9

3.1. Remittances and Labour force Participation………………………………..9

3.2. Post-communist migration..……………………………………………….11

3.3. Cultural context…………………………………………………………...13

4. Data…………………………………………………………………………13

5. Methodology………………………………………………………………..18

6. Estimations…………………………………………………………………21

7. Instrumental Variable Approach…………………………………………27

7.1. The Instruments………………………………………………………….. 27

7.2. First Stage Results………………………………………………………...29

7.3. Reduced Form Estimates………………………………………………….30

7.4. Second Stage Results……………………………………………………...32

8. Problems with the exchange rate instrument……………………………34

9. Conclusions…………………………………………………………………38

10. Bibliography………………………………………………………………40

11. Appendix…………………………………………………………………..43

  4  

List of Figures and Tables

Figure 1. Remittance inflows in Albania, years 2002-2014………………...9

Figure 2. Remittances/GDP ratio……………………………………………10

Figure 3. Labour Force Participation (percentage)………………………...11

Table 1. Sampling in terms of Districts……………………………………..14

Table 2. Summary Statistics…………………………………………………15

Table 3. Pooled OLS of Remittances in Unwillingness to work……...……23

Table 4. Pooled OLS of Number of Household Members in unpaid job.....25

Table 5. Pooled OLS of Remittances in Unwillingness to work, Restricted

Sample………………………………………………………………………...26

Table 6. First Stage Results………………………………………………….29

Table 7. Reduced Form Results……………………………………………..31

Table 8. Second Stage Results (No PSU dummies)………………………...33

Table 9. Second Stage Results (with PSU dummies)……………………….35

Table 10. Correlation between Remittances and Exchange rate………….37

Table 1. Appendix, Hausman Test for endogeneity………………………...43

  5  

1.Introduction

The purpose of this study is to investigate in the impact of remittances in the (un)willingness

to work in Albania. There are several reasons behind the choice of this issue as a research

question. First of all, remittances are one of the main components of the GDP for many

developing and emerging economies. World Bank statistics show that in 2014 the global

estimated remittance flows were $580 billion, and the number of international migrants in 2013

was 247 million (World Bank, 2017). Considering that migration and remittances are an

important factor in many of developing and emerging countries’ economies, many researchers

find it useful to study the impact that remittance inflows have in these countries, in both

microeconomic and macroeconomic terms. As it will further be presented in the literature

review section, there is a broad discussion among the researchers regarding the impacts that

remittances have in different aspects of economic indicators and development, in the

remittance-receiving countries. Regarding the effects that remittances have in the labour supply

decisions in the receiving countries, these studies suggest that in most of the cases remittances

are associated with a decrease of the labour supply, and increase of the reservation wages.

Based on these results, one can expect that remittances will affect positively the unwillingness

to work, by decreasing working incentives of the remittance-receiving households.

The first reason why I use Albania as the country on which the research question will be

developed, is due to the data availability for Albania. I use micro data from the Household

Budget Survey (HBS) for years 2009 and 2014, which are obtained in collaboration with the

Institute of Statistics of Albania, and are not available online. Another reason why I consider

this research question to be applicable for the case of Albania is that Albania is one of the

countries that has faced high migration rates since after the fall of communist system, and

nowadays remittances play a significant role in the country’s economy. This is primary

reflected by the share of remittances in the GDP over years. The average share of remittances

on GDP form years 2005 to 2014 is 8,91% (Bank of Albania, 2017). The considerable share

that remittances have on Albania’s GDP, raise some natural concerns regarding the impact that

remittances have on different aspects of economic development.

In this study, I aim to investigate the impact that remittance inflows have on the unwillingness

to work for the remittance-receiving individuals. The starting point of the choice of this

research question are the high unemployment rates that Albania has been facing over years.

  6  

These unemployment rates during the time period 2007- 2016, have varied from 13,2% to

17,9% (INSTAT, 2017). These high rates of unemployment are an indicator of a problematic

labour market and unhealthy economy. Therefore, taking these numbers into account, and also

the fact that many households have remittances as an additional source of income, it is

interesting to observe whether remittances affect in the increase of decrease of the

(un)willingness to work for the individuals belonging to these households. The unique feature

of the research question developed in this study, is the fact that unlike most of the previous

studies, here is investigated the impact of remittances in work incentives (unwillingness to

work), instead of the participation in labour market.

In order to assess the effects of remittances in the unwillingness to work, I use pooled OLS

with and without controlling for month and Primary Sampling Units fixed effects, and further

proceed with an Instrumental Variable approach, as an attempt to find a causal effect and avoid

endogeneity problems and potential omitted variable bias. The results of the pooled OLS

indicate that an increase of the amount of remittances with one standard deviation, is associated

with an increase with 0.009 units of the unwillingness to work. This coefficient is robust and

statistically significant after the inclusion of month and PSU dummies. Also, this coefficient is

robust while estimating the restricted sample, and the magnitude is slightly decreased in 0.008.

In the case of IV estimation, the magnitude of the coefficient (while using strong instruments)

is higher and reaches the value of 0.026. However, this coefficient is not robust after the

inclusion of PSU dummies in the regression. The insignificance of the coefficient in the later

case, may be related with possible problems that are related to the exchange rate instrument.

However, despite the limitations of this study, the results indicate that remittances positively

affect the unwillingness to work, by reducing working incentives.

This paper is organized as following: section 2 presents the literature review, section 3 provides

background information about Albania in general, as well as emigration and remittances in

Albania, section 4 describes the data, section 5 presents the methodology used, section 6 shows

the results and their interpretations, section 7 elaborates on limitations and possible

explanations about the problems, and section 8 concludes.

  7  

2. Literature Review

In general, there cannot be achieved a broad consistency among the researchers on a certain

positive or negative impact that remittance inflows have on the remittance-receiving countries.

There are studies that show supportive evidence on the positive impact of remittances in the

receiving countries. Usually they find that remittances are helpful in decreasing poverty and

increasing development of the receiving countries.

However, several studies have concluded that besides the positive effects, some negative

impacts are also present in both micro and macro terms for the receiving countries. The studies

that highlight the positive effects of remittances, find these effects mainly in terms of poverty

alleviation and increasing development mostly in developing and emerging economies. For

instance, (Gupta, Pattillo, & Wagh, 2007) show empirically that remittances are helpful to

alleviate poverty and promote financial development in sub-Saharan Africa. In a macro-

perspective, (Giuliano & Ruiz-Arranz, 2005) use a cross-country dataset of 30 developing

economies for the time period 1975-2002 and find that in less countries that were less

financially developed, remittances promoted growth. In their study for six remittance-receiver

countries including Albania, (Meyer & Shera, 2016) use a fixed effects approach and find that

remittances positively affect GDP growth.

On the other hand, there are studies that confirm some negative aspects of remittance inflows

in the receiving countries. These are related to the exchange rate appreciation, increase of the

non-tradable sector and consumption of foreign goods. Moreover, evidences presented in some

studies suggest decrease on the labour supply of the remittance-receiver households, who tend

to increase leisure, decrease labour, and increase their reservation wages, given remittances are

a reliable additional source of income for them. (Acosta, Lartey, & Mandelman, 2007) show

for the case of El Salvador that remittance inflows are accompanied with the Dutch disease,

which is associated with exchange rate appreciation and an increase of the non-tradable sector,

relative to the tradable sector. Furthermore, they show that remittance inflows decrease labour

supply. In a study for Mexico, (Airola, 2008) finds that remittances affect labour supply

participation decisions, by reducing the hours worked of the head of the households. This

magnitude of this effect is larger for females. The same effect is also found in the paper of

(Kim, 2007) ,who investigates the case of Jamaica, a remittance-receiver country and faces

high unemployment rates over years. The households who receive remittances are found to

  8  

have higher reservation wages, which affects their labour participation decisions. In their study

for temporary migration in case of Philippines, (Rodriguez & Tiongson, 2001) find that

migration reduces the labour supply of non-migrant family members, and the effects vary

across gender, differences in education, and family ties among the households.

Other studies investigate the impact of migration and remittances particularly for Albania.

(Konica & Filer, 2009) use survey data from year 1996 and observe the link between

remittances and starting an individual or family enterprises, and the effect of remittances

received by the households in their labour force participation. Their findings suggest that

remittances supported the development of private enterprises, and reduced labour supply of

females. Another early study by (Gedeshi, 2002) presents descriptive statistics based on the

responses of the interviewed Albanian emigrants mainly in the neighbouring countries. One of

the main findings of this study is that remittances are mainly used for “financing the immediate

necessities of the family in the homeland”, and a much smaller share of the respondents aimed

to use remittances as a “source for financing their own investment in Albania.” This indicates

that remittances played a major role in reducing poverty and improving living conditions, rather

than being used for investment purposes. Furthermore, the author elaborates on the trends of

remittance inflows and the factors that influence these trends, but not all the factors can be

related to the actual environment, since there have been many changes in the economic and

social environment during the years. In a more recent empirical study, (Dermendzhieva, 2010)

investigates on the impact that remittances have in labour supply in Albania. The author uses

survey data from the Living Standards Measurement Survey 2005 in Albania, and controls for

the endogeneity issues by using Instrumental Variables. The IV results of this study suggest

that remittances have a negative and significant effect in the probability of working for married

females, and males in the groupage 46-60 years old.

Overall, these studies suggest that remittances are helpful on alleviating poverty and

developing some sectors of the economy, such as small household enterprises, but there should

be paid attention to the negative impacts as well, in order to be able to design effective policies

that may help to decrease the negative effects.

  9  

3.Backgruond information

3.1. Remittances and labour force participation

According to World Bank Development Indicators, in terms of income, Albania is classified

as an upper-middle income country (World Bank, 2016). In terms of the World Economic

Situation and Prospects classifications, Albania is classified as economy in transition (United

Nations, 2014). By definition, a transition economy is known as “an economy that is changing

from being one under government control to being a market economy (Cambridge Dictionary,

2017).” Starting from the early 1990s, remittances have been and continue to be a reliable

source of income for many households in Albania. A large fraction of people has migrated

during this period mostly in neighbouring EU countries such as Greece and Italy and has

continuously sent money to their families and relatives back home. In a macroeconomic

perspective, this is reflected by the share of remittances on GDP, which has varied from 12 to

5,64 percent of the GDP on the last ten years (Bank of Albania, 2017).

Considering these numbers, remittances appear to be an important component of the GDP of

Albania, implying that they might play a significant role on the budget of many households. In

figure 1. there are presented the remittance inflows in Albania (in million Euros) during the

time period 2002-2014. As shown in the figure, the total amount of remittance inflows reached

the maximum values in 2007 with the total amount of remittances of 951.2 million Euros,

whereas the minimum was in 2013 with the corresponding value of 543.8 million Euros.

Figure 1. The y-axis represents the amount of remittances flows in Albania (million Euro) and the x-

axis represents the years. Data Source: Bank of Albania, author’s presentation.

693  

774 774802

937 951.7

833.3781.3

689.8665 675

543.8591.9

2002 2003 2004 2005 2006 2007 2008 2009 2010 2011 2012 2013 2014

Remittances  in  Albania  years  2002-­‐2014  

  10  

 Figure  2.  The  y-­‐axis  represents  the  ratio  of  remittance  inflows  over  GDP  in  Albania  (in  percentage).  

The  x-­‐axis  represents  the  years.  Data  Source:  Bank  of  Albania,  author’s  calculations.  

Besides the presentation of the amount of remittance inflows in Albania, it is interesting to

observe how these numbers are reflected relative to GDP. Therefore, in figure 2., there are

presented the ratios of remittances over GDP from year 2002 to 2014, expressed in percentage.

In contrast with the previous figure, from 2003 the trend of the ratio is decreasing up to 2005.

In 2006 remittances account for 13,1% of GDP, and this ratio continues to decrease and reaches

the minimum value in 2013, which is 5,6% of GDP. In 2014, the ratio starts to increase,

following the same trend as the amount of remittance inflows presented in figure 1. The data

presented in this figure demonstrate that remittance inflows not only are present as high

amounts of monetary inflows that Albanian economy receives every year, but also they account

for a considerable share of GDP, with a minimum percentage of 5,6% and a maximum of

15,3% of GDP.

The other side of the research question is the unwillingness to work, which, in macroeconomic

terms can be reflected by the participation in labour force. Figure 3. presents the labour force

participation rates for years 2007-2016 (INSTAT, 2017). The data presented in the graph show

that there is a high volatility on the labour force participation rate from one year to another. An

example of this volatility is the difference between the labour force participation rates in two

subsequent years: 2010 and 2011 where this rate has been increased with 6%. Also, from 2012

to 2013 the labour force participation rate has been decreased with 5,6%. The fact that the

14.7 15.313.2 12.2 13.1 12.2

9.4 97.6 7.1 7.04

5.6 5.95

2002 2003 2004 2005 2006 2007 2008 2009 2010 2011 2012 2013 2014

Ratio  Remittances/GDP  in  percentage

Ratio  Remittances/GDP  in  percentage

  11  

labour force participation rates have the tendency to incur considerable changes even in short

periods of time, indicates that labour force participation can be easily influenced by factors that

are directly related to the (un)willingness to work. On the other hand, considering that

remittances account for a considerable share of Albania’s GDP over years, it is interesting to

observe whether the amount of remittances received by the households affects their

(un)willingness to work, which consequently affects the participation on the labour force.

Before proceeding with data, methodology, and empirical findings, in the next subsections I

will briefly elaborate on historical and cultural factors that have affected international

migration and remittance inflows in Albania.

Figure 3.: Labour force participation rate in percentage. X-axis represents the years, Y-axis represents

the percentage of the labour force participation rate, data source: INSTAT, author’s presentation

3.2. Post-communist migration

The communist regime took place in Albania for almost 50 years from 1945 to 1990. One of

the features of the regime was the prohibition to travel, migrate, and even become influenced

by the western culture. After the fall of the communist regime, Albania was facing the

challenges of transition from a centralised economy, to an open market economy, which was

accompanied by poverty, unemployment and many other economic and social difficulties.

Therefore, influenced by the western media, and eager to seek for a prosperous life Albanians

65.2

61.9 61.9 62.2

68.2

65.5

59.9

61.5

64.2

66.2

54

56

58

60

62

64

66

68

70

2007 2008 2009 2010 2011 2012 2013 2014 2015 2016

Labour  Force  Participation  (percentage)

Labour  force  participation

  12  

massively migrated mainly in the neighbouring countries such as Greece and Italy, taking

advantage from the geographical proximity of these countries with Albania (Carlo Azzarri,

Calogero Corletto, 2009).

The first episodes of post-communist migration were in 1990 and 1991. On 1990 around 5000

Albanians who wanted to leave Albania, had the refugee status through the Embassies of some

of the Western European Countries. Afterwards, almost one year later, a large number of nearly

26000 people who desperately wanted to live and work abroad, went in Italy. They arrived in

Italy in overcrowded ships via the Adriatic Sea, and the Italian government allowed them to

reside there by giving them the refugee status. Meanwhile, for many those who did not choose

migration as a way to improve their lives, things got deteriorated with the arise of the “pyramid

saving schemes”. Taking advantage from the fact that most of the people were not adequately

informed regarding the way that the market economy, financial markets and investment

schemes worked, the entrepreneurs that created these schemes deceived the Albanians. They

claimed that saving money in these schemes would allow people to receive very high interest

rates on their savings. The interest rates reached extremely high values in 1996 up to 200% in

the moths prior to their collapse. Being attracted by the very high interest rates, approximately

2 million people put their savings (many of them even sold their houses or properties) in these

defrauding schemes. The total amount of these savings was around 2 billion dollars. After more

than three years of operation, these firms collapsed and all the people who had their deposits

in these firms lost their money. This created a chaotic situation inside the country, leading to

economic and political instability accompanied by civil conflicts (Jarvis, 1999). Under this

desperate situation, many people saw migration as a way to resolve their problems and escape

poverty. Almost 70000 Albanians migrated in that period. The migration flows were present

even in the subsequent years (Bajraba & Bajraba, 2015).

According to the United Nations Statistics’ on International Migration, until the year 2000

there were 822,676 migrants whose origin country was Albania. These numbers have continued

to increase reaching 1,122,910 in 2015 (United Nations, 2016). However, regardless the fact

that the numbers of migration flows are not as high as during the early 2000s, migration is a

continuing event in Albania, since many people view migration as a golden opportunity to

improve their lives and their families’ lives. Most importantly, the tendencies to migrate are

present among the young generation, meaning that this trend will continue to be present in the

  13  

Albanian society. Survey statistics of 2015 on young people confirm this reality by showing

that 59.8% of the respondents aspire to migrate abroad (Çela, Kamberi, & Pici, 2015).

3.3. Cultural context

The model of the typical Albanian family can be described as a consolidated and traditional

family, where people are closely linked to each other and all the members tend to contribute in

the family’s wellbeing. Even though the tendency to embrace a more modern mentality

regarding familiar and social relationships is increasing, the role of family continues to remain

important in many aspects of peoples’ lives and decision making. Not only the earlier

generations, but also the young people nowadays, are closely linked to their parents and/or

siblings. Most of them consider family members as the people to whom they can rely mostly

while facing difficulties. Furthermore, they look at the family’s influence in their lives and

decisions as a positive element. For instance, the majority of young people, despite their

economic background prefer to live with their parents rather than on their own, confirming that

the foundations of the traditional family continue to remain present (Çela, Kamberi, & Pici,

2015).

The presence of the traditional elements in Albanian families, suggests that family members

besides potential benefits, have also responsibilities towards each other. In the case of people

who live and work abroad, this is translated into a moral obligation to provide financial

assistance to their families and/or relatives in their homeland. According to the ‘Questionnaire

on the foreign currency remittances of long-term legal immigrants’ conducted from the Bank

of Albania in 2003, the main recipients of remittances were the parents, wives and children of

the migrants. The main purpose of sending money to their families resulted to be: ‘meeting the

essential needs of the family’, followed by ‘furnishing the house’ (Uruçi & Gedeshi, 2004).

4. Data

The data are collected in collaboration with INSTAT (Institute of Statistics in Albania), which

is the official institution for data collection and analysis in Albania. These data are part of the

datasets of Household Budget Survey(HBS), and are not available online. The data is collected

by interviewing households regarding their income, expenditures, employment status and other

individual characteristics, such as age and education. The dataset contains data from 2009 and

2014 and the total number of observations is 42426. For reasons that will be explained further,

  14  

I use only the observations for the head of the households in the regression analysis. The total

number of observations for the head of the households is 12160, whereas the number of

observations for 2009 is 5596 and the number of observations for 2014 is 6565. The sample

selection is random, meaning that the respondents are from all of the main regions in Albania,

from both urban and rural areas. There are 12 districts in Albania, and from each district there

have been interviewed different households. The number of observations is higher in the

districts with a higher population density. For example, in Tirana district, which includes the

capital city Tirana and has the highest population density, the number of observations is 2867,

whereas in Kukës, which has the lowest population density, the number of observations is 546.

Table 1. presents the number and the percentage of observations for each district.

 Table  1:  Sampling  in  terms  of  Districts  

District  Name     Frequency   Percentage  Berat   703   5.78  Dibër   553   4.55  

Durrës   1044   8.59  Elbasan   1410   11.60  

Fier   1460   12.01  Gjirokastër   472   3.88  

Korçë   990   8.14  Kukës   315   2.59  Lezhë   546   4.49  

Shkodër   1035   8.51  Tiranë   2,867         23.58        Vlorë      765       6.29  Total   12,160         100.00  

   As shown in the table, the highest percentage of observations is from Tirana, the district of the

capital city, which is the most populated city in Albania, followed by Elbasan, Fier and Durrës.

Furthermore, the data are arranged in terms of smaller groups, based on the geographic

proximity of interviewed households from each other, called Primary Sampling Units. In total

there are 653 primary sampling units that contain observations for each household.

Up to this point, the quality of the data is good given the randomness of the sample and the

availability of the time variation. However, the observations are not from the same households

in both years. Unfortunately, this does not allow for a panel data analysis. The availability of

panel data would allow me to use fixed or random effects estimations, which would be very

helpful to solve the omitted variables problem, and come closer to the estimation of a causal

  15  

effect. Given that this is not possible due to the nature of the data, I use a time dummy to control

for the time variation, and use month fixed effects and primary sampling unit fixed effects. I

prefer primary sampling unit fixed effects to district fixed effects because of the smaller size

and largest number of the primary sampling units. Then, I proceed with the analysis by using

the indicators for the head of the household. I use the observations for the head of the

households only, because of the inability to control for within-groups(households) fixed

effects. In Table 2. there are presented summary statistics about the variables included in the

regression.

Table  2.  Summary  statistics  for  the  variables  of  interest    Variable   Observations   Mean   Standard  

Deviation  Minimum  Value  

Maximum  Value  

 Yearly  Amount  of  Remittances  

12160   1720259   8548071   0   421000000  

Unwillingness  to  work  (0=willing  to  work,  1=not  willing  to  work)  

12160   .0133557   .0656835   0   1  

Age  of  the  Head  of  the  Household    

12160      55.72961         13.31466         12   101  

 School  Years  of  the  Head  of  the  Household  

12160   9.605345         3.749815   0   22  

Head  of  the  Household  is  Female  (0=Male,  1=Female)    

12160   .1372533   .3441287       0   1  

Head  of  the  household  Married  (0=Single,  1=Married)  

12160   .3787007   .4850833   0   1  

Number  of  Persons  in  the  Household  

12160   3.890296       1.715846   1   19  

Total  consumption  expenditure  (quarterly)    

42426   739542.7   522074.9   28000   8776090  

Total  annual    income   42426      4352559   17100000   0   1510000000      

Table  2:  The  amounts  are  expressed  in  Albanian  currency:  old  leks.1  Euro=1350  old  leks  

  16  

The main variable of interest is the amount of remittances that each household receives. This

represents the yearly amount of remittances received expressed in old leks (Albanian currency).

The interviewers have declared the amount of remittances that they have received in the last

12 months. The mean of this variable is 1720259 old leks, which is approximately equivalent

to 1274 Euros. This implies that in average, a remittance-receiver family receives 109 (14715)

euros(lek) per month. In terms of the minimum wage in Albania, which is 23000 lek per month,

the average remittance incomes exceed half of a monthly minimum wage.

However, there exists the possibility that the amount of remittances is underreported mostly by

the households who receive high amounts of remittances. One of the explanations regarding

this phenomenon is the “fear” of the households to declare that they receive high amounts of

remittances, given that in most of the cases they are not being sent in formal channels i.e. banks,

but they are transferred as cash money by the emigrant him/herself, or by their relative or

friends towards their families. Unfortunately, the existence or magnitude of this phenomenon

cannot be controlled, since one cannot control whether the households are truly declaring the

amount of their remittance incomes. Yet, I take into account the fact that this problem may be

present, and restrict the sample by excluding the highest remittance-receiver households, as a

robustness check for the analysis in the next sections.

For the main explanatory variable “remittances” there are three possible forms of inclusion in

the regression. The first one is in its binary form that shows whether the individual receives or

not remittances. The second one is by including the amount of remittances that each individual

receives, and the third one is the standardized amount of remittances received. The preferred

alternative is the standardized amount of remittances because it is based on the amount of

remittances received and the beta coefficient of the standardized values represents the impact

of an increase of the amount of remittances with one standard deviation. The standardization

of the variable can be interpreted as a way to make the scaling of the units negligible, and thus

allows for an easier interpretation of the magnitude of the coefficients. The standardized values

of a variable can be obtained by subtracting the mean of the variable from their each of the

values, and dividing the values by the standard deviation (Wooldridge, Introductory

Econometrics, A Modern Approach, 2013).

The dependent variable in the regression that will be presented further is the unwillingness to

work and takes the values between 0 and 1. It indicates the proportion of the persons that are

  17  

not willing to work within a household. The minimum value of this variable shows that none

of the persons in the household is unwilling to work, whereas the maximum shows that all the

persons are not willing to work. For constructing this measure, the respondents were asked

whether they were willing to work if they were offered a job within 2 weeks form the moment

they were interviewed. Although this seems to be a good proxy for measuring the individuals’

unwillingness to work, there are two potential problems related to this measure. The first one

is the potential measurement error that arises due to time inconsistency of the responses

regarding the willingness to work. The respondents have provided their answers based on the

time when they were interviewed. For instance, a responded who has been interviewed in

January has been answered to the question based on the following two weeks of January,

whereas a respondent who has been interviewed in June is based on the following two weeks

of June for answering to the question. If the respondents’ answers would have been affected

by the time on which they provided their answers, then the measurement error would be a

serious issue in that case. However, one cannot control whether the respondents’ answers

would have been different if they answered in a different time of the year. Therefore,

considering that all the respondents were interviewed within one year, it is very likely that their

answers would not change based on the month of the interview, which minimizes the issue of

measurement error. The second problem is related to the way that people tend to answer to this

question: some of them may feel uncomfortable to report that they would not be willing to

work if they were offered a job. Some of the reasons behind this may be: the mistrust to the

interviewer regarding the confidentiality of their answer, the fear of being prejudiced by the

interviewer, and also the fear that other people or even government authorities may be informed

about their answers. Unfortunately, one cannot be able to control the truthfulness of the answers

of the respondents. Overall, despite the two potential problems that were mentioned regarding

the measure of the (un)willingness to work, in terms of this dataset, this is the best proxy that

can be used in order to provide an accurate measure of the (un)willingness to work.

The variable “Head Age” indicates the age of the head of the household. Therefore, the mean

of this variable is 55.7, which indicates a relatively old age. The variable “Head School years”

presents the number of school years for the head of the household. The mean of this variable

9.6, and the minimum and maximum values range between 0 and 22 years of schooling. The

variables “Head Female” and “Head Married” are binary variables that indicate whether the

head of the household is a female or married. The last variable presented in Table 2., presents

the number of the persons in a household. The mean of this variable is 3.8, indicating that in

  18  

average a family is composed by 4 persons, and the minimum and the maximum number of the

persons within a household range from 1 to 19.

In the last two rows of the table there are shown summary statistics about total quarterly

spending and total annual income. I use the total number of observations while presenting these

statistics in order to create a clearer idea regarding income and expenditure levels. The mean

of the total quarterly expenditures is 739542.7 old leks. The minimum value of total quarterly

expenditures is 28000 old leks, and the maximum value is 8776090 old leks. By comparing the

average amount of consumption with the average amount of remittances in quarterly basis (for

the remittance-receiving households), it results that remittances may account for more than a

half of consumption expenditures. In terms of the research question, these numbers show that

remittance-receiver households are more likely to not prefer to work, given that they can cover

more than a half of their consumption expenditures through remittances. The mean of the

annual income level is 4352559 old leks, whereas the minimum income level is 0 and the

maximum level of annual income is 1510000000 old leks. The minimal and maximal values

of yearly income indicate that in the sample there are observations that indicate for unemployed

people, whose income level is 0, and also rich people, whose income levels are extremely high

compared to the average income level.

5. Methodology

This section describes the methodology used for addressing the research question. The first

step for estimating the relationship between remittances and unwillingness to work, is the

construction of the baseline equation. In this regression the dependent variable is the

unwillingness to work, which shows the fraction of the individuals that are not willing to work

within a household. The main explanatory variable is the standardized amount of remittances.

As mentioned in the data section, this measure is preferred towards the amount of remittances

because it allows to avoid the problems that arise as a result of the scaling units. Equation (1)

presents the baseline regression:

(1)  𝑈𝑛𝑤𝑖𝑙𝑙𝑖𝑛𝑔𝑒𝑠𝑠)* = 𝛽-𝑅𝑒𝑚𝑖𝑡𝑡𝑎𝑛𝑐𝑒𝑠)* + 𝛽4𝑋)*Γ + 𝑌𝑒𝑎𝑟49-: + 𝜀)*

The second term in the right hand-side of the equation represents a vector of control variables

included in the regression. The control variables indicate the Age of the head of the household,

  19  

the quadratic term of age, marital status of the head of the household, gender of the head of the

household, number of school years of the head of the household, and the number of the people

living within a household. The third term on the right hand-side of the regression represents a

time dummy that takes the value of 0 if the year is 2009, and the value of 1 if the year is 2014.

The term 𝜀)* is used to represent the unobserved factors that remain in the error term of the

regression. Further in this section, this error term will be decomposed in two parts, in order to

demonstrate the distinction between the variables that may be used as instruments, and the

other unobserved variables. The baseline equation is first estimated by using pooled OLS,

without fixed effects, where observations of both years are included. Next, I include Month

dummies and Primary Sampling Unit dummies, as a way to capture month fixed effects and

primary sampling unit fixed effects.

The reason behind using these dummies for capturing the fixed effects, is the intention to

minimize the omitted variables bias problem. The presence of the omitted variable bias

problem can be attributed to the fact that usually it is not possible to include all the explanatory

variables in the regression. Therefore, due to data availability, or measurement problems, many

of the potential explanatory variables that are related to the dependent variable, remain in the

error term. In case of this regression, one of the possible omitted variables is the reservation

wages: the reservation wages are very likely to be positively correlated with the unwillingness

to work. The explanation about this correlation is intuitive: as the reservation wage of an

individual increases, his/her unwillingness to work will increase as well, given that he/she will

require a higher wage in order to participate in the labour market.

After using pooled OLS and controlling for month and PSU fixed effects, the final step towards

overcoming the endogeneity problems, and finding a causal effect, is the Instrumental Variable

approach. Before elaborating further on the challenges that are associated with the IV approach,

I will first investigate whether the variable of remittances is endogenous, which would make

IV necessary. In order to do this, I run a Hausman test for endogeneity. This test is used to

show whether the differences between OLS and IV estimates are statistically significant, and

if this is the case, one can conclude that the variable of interest, in this case remittances, is

endogenous (Wooldridge, Introductory Econometrics, A Modern Approach, 2009). Table 1.,

in the Appendix shows the results of the Hausman test for endogeneity. The results of this test

show that the null hypothesis which states that the differences in coefficients are not systematic,

  20  

is rejected at 1% level. This implies that the variable of remittances is endogenous, and

therefore it is necessary to use an IV.

The main challenge of using the IV method is to find an instrument that is exogenous and

related to the dependent variable “not willing to work” only through remittances. Geographical

indicators such as: distance from a certain point, are considered to be good instruments by the

researchers, since they seem to satisfy both the exogeneity condition, and exclusion restriction.

However, there cannot be found a general solution regarding the choice of the instruments.

This choice of the instrument depends on the specific regression equation that the researcher is

estimating. For each specific equation that one is interested to estimate the instrument needs to

be exogenous and affect the dependent variable only through the instrumented variable.

Finding a good instrument is challenging, not only because there has to be satisfied the

exogeneity condition, which should not be confounded with the externality of the variable

chosen as an instrument, but also because this condition cannot be tested empirically (Deaton,

2009).

The general form of the regression can be presented as:

(2)      𝑈𝑛𝑤𝑖𝑙𝑙𝑖𝑛𝑔𝑒𝑠𝑠)* = 𝛽-𝑅𝑒𝑚𝑖𝑡𝑡𝑎𝑛𝑐𝑒𝑠)* + 𝛽4𝑋)*Γ + 𝑌𝑒𝑎𝑟49-: + 𝑢)*+ 𝜇)*

The notation of the variables is the same as in equation (1). The only difference here is in the

decomposition of the error term, where 𝑢)* represents the exogenous variables that are not

included in the regression and can potentially be used as instruments, and 𝜇)* represents the

remaining unobserved factors.

The main purpose in this research is to find a casual effect of remittances in the unwillingness

to work, and the main threat to causality is the possible endogeneity of the variable of

remittances. Therefore, a possible solution is to find an instrumental variable, Z that affects

only the amount of remittances, but not the error term. Mathematically, these conditions can

be expressed as:

1.   𝐸[𝑋|𝑍] ≠ 0

2.   𝐸 𝜀 𝑍 = 0

  21  

The Instrumental Variable approach consist in two stages. In the first stage the instrumented

variable “amount of remittances” is regressed on the instrument. This is a way to exploit the

exogenous part of the main explanatory variable. The regression can be presented as:

(3)  X=Z𝛾 + 𝜌,

where X is the amount of remittances, Z is the instrument, 𝛾  is the coefficient of the instrument,

and 𝜌 is the error term. The main conditions for the first stage to be valid are the F-statistic

above 10, and 𝛾 statistically significant. Then, after filtering the exogenous part of the

instrumented variable in the first stage, this part is used in the second stage regression, which

is the regression that presents the impact of the X variable on Y.

In the context of exploiting the causal relationship between remittances and willingness to

work, I use the distance from land borders and the distance from the two ports of each

Prefecture in Albania to construct the geographical instrument. The fact that the data are

constructed based on 12 Prefectures allows me to assign the average distance from each

Prefecture to the closest land border, to each observation based on their location. The same

logic is followed for constructing the sea distance instrument: given that there are two ports

from where individuals can migrate, I estimated the average distance of each Prefecture to the

closest sea port. Then, after obtaining separately the variables that represent land distance and

sea distance, I construct the total distance instrument as the sum of land and sea distances.

Another variable that is related to the dependent variable only through remittances, although

as will discussed below there may be other channels of correlation that may question this, is

the exchange rate. Since the observations correspond to different months, the average monthly

exchange rate for each year can be assigned to each observation, giving a new variable that

represents the exchange rate. This variable can be used as an instrument separately, or as an

interaction with the total distance. The results are presented in the next section, and each of the

instruments and the outcome will be discussed in more details below.

6. Estimations

In this section there will be presented the estimation results for the baseline regression, first

without including month and primary sampling unit fixed effects, and then with their inclusion,

both separately and simultaneously. Table 3., presents the results of the baseline equation.

  22  

Column (1) shows the results of the simple OLS regression, where the only explanatory

variable is the standardized amount of remittances. The coefficient is positive and significant

at 1% level, meaning that this coefficient is significant in statistical terms. In an economic

perspective, the magnitude implies that an increase in the amount of remittances with one

standard deviation, is associated with an increase of the unwillingness to work with 0.0099

standard deviations. While using only the standardized amount of remittances as a control

variable, the R-squared is 0.0227.

Column (2) presents the results of the multiple regression, which is based on the baseline

equation (1). As in the previous case, the coefficient of remittances is statistically significant

at 1% level, and its magnitude has been slightly decreased compared to the coefficient of the

simple regression. The coefficients of head age and the quadratic term of head age, are both

statistically significant at 1% level.

The coefficient for head age has a positive sign, implying that as head age increases, the

unwillingness to work increases with 0.0015 units, whereas the coefficient of head age squared

has a negative sign. This can be related to the trade-off between education and work: while the

individual invests in education he/she is more likely to reduce his/her work incentives. Then,

at a further moment in life, the individual increases his/her work incentives, given that he/she

has already invested in education and professional skill improvements. Therefore, in that point

of time, the individual does not need to choose between working and schooling, since working

is the best option in that moment.

The estimated coefficients of head married and head female are not statistically significant in

this regression, even though the sign is negative. The coefficient for head school is significant

at 5% level, and it is negative. The interpretation for the magnitude of this coefficient is that

an increase with one year of schooling for the head of the household is associated with a

decrease by 0.0004 units of the unwillingness to work. Thus, an increase in education is

associated with a decrease of the work disincentives. The coefficient for the number of persons

in the household is positive and significant at 10% level. This may be attributed to the fact that

a larger number of members in the family is associated with more household duties such as:

taking care of the children and maintaining the dwelling. Therefore, the work incentives are

likely to be reduced in this case.

  23  

Table  3.  Effect  of  remittances  in  the  unwillingness  to  work,  Pooled  OLS     (1)   (2)   (3)   (4)  

Remittances      

Head  Age      

Head  Age2      

Head  Married  

.0099***  (.00058)  

.0089***  (.0006)    .00149***  (.0003)        -­‐.000013***  (.000002)      -­‐.0037      (.0027)  

.0090  ***  (.0006)    .0015***  (.0003)    -­‐.0000133***  (.000002)      -­‐.0037    (.0027)  

.0089***  (.0025)    .0014***  (.0003)    -­‐.0000119***  (.000002)      -­‐.0030    (.0042)  

                 Head  Female     -­‐.00063   -­‐.00068       -­‐.00089  

    (.00208)    

(.00208)   (.0026)  

Head  School      -­‐.000419**  (.00017)  

-­‐.00042**  (.00017)  

-­‐.00008    (.0002)  

         HH  Number  

   

Year  2014      

                       Constant  

             .01335***  (.00058)  

.00068*  (.00036)    -­‐.01293***  (.0025)      -­‐.00358    (.00973)  

.00066*  (.00036)    -­‐.01290***  (.0025)        -­‐.00293    (.00992)  

.00038  (.00038)    -­‐.01032**  (.0039)      .00288  (.010104)  

         Month  Fixed  

Effects    

PSU  Fixed  Effects  

 No    No  

 No    No  

 Yes    No  

 Yes    Yes  

         N  

R-­‐Squared  F-­‐Statistics  

12160  0.0227  283.69  

12160  0.0307  48.08  

12160  0.0318  22.00  

12160  0.0862  

Standard  errors  in  parenthesis,  *  significant  at  10%  level,  **  significant  at  5%level,  ***significant  at  1%  level,  no  stars  implies  not  significant.    

In order to demonstrate this correlation, Table 4., presents the results of the simple and multiple

regression of unpaid job to number of household members. The unpaid job is referred to

working in an individual farm, or maintaining the dwelling. The coefficient of HH number in

the simple regression presented in Table 4., column (1), indicates that there is a positive and

  24  

statistically significant correlation between the number of family members within a household,

and working in an unpaid job. In columns 2 and 3 there are presented the results of multiple

regressions of unpaid job in the number of people in the household, with and without including

Month and PSU dummies. As it can be inferred from the results presented in these columns,

the coefficient of the number of household members remains statistically significant at 1% in

column 2 and 5% in column 3, and its sign is positive. The magnitude of this coefficient

decreases as more control variables are included in the regression. Overall, the results of Table

4., indicate that there is a positive and significant correlation between the number of household

members and working in an unpaid job, which is robust after including additional control

variables and Month and PSU dummies.

In Table 3., the coefficient of the year 2014 dummy is negative and significant at 1% level. The

R-squared is now 0.0307, which is higher compared to the simple regression. In column (3)

there are presented the results of the baseline regression after adding month dummies. The

coefficient of remittances continues to be statistically significant at 1%, and its magnitude has

been slightly increased compared to the results in column 2. The same pattern can be observed

for the coefficients of the other explanatory variables as well: their statistical significance is

the same as in the previous results, and their magnitude is associated with minor changes. The

R-squared has also been slightly increased.

Column (4) in Table 3., shows the results of the baseline regression with month fixed effects

and primary sampling unit fixed effects. The coefficient of remittances has been slightly

decreased, but continues to remain statistically significant at 1% level. The coefficients of head

age and head age squared also show the same pattern: they remain significant at 1% level, and

show minor changes in their magnitude. The difference with the previous results can be found

in the coefficients of HH number and head school, which became insignificant. Also, the

coefficient of year 2014 is no longer significant at 1% level, but at 5% level, although the sign

is the same. The R-squared has been increased and has reached the magnitude of 0.0862.

Overall, based on the results obtained so far, the main finding is that the coefficient of

remittances remains robust, despite different specifications used for the baseline equation. The

inclusion of the month dummies and primary sampling unit dummies, did not change neither

the significance nor the sign of the coefficient. Also, the magnitude has shown only minor

changes in all the different specifications.

  25  

Table  4:  Regression  of  working  in  an  unpaid  job,  main  explanatory  variable:  number  of  people  in  a  household  

  (1)   (2)   (3)  Number  of  HH  

members    

Remittances  

.0077***  (.0007)    

.0035***  (.0007)    -­‐.0017  

.0018**  (.00073)    -­‐.0031***  

                                                                                   Age  of  Head  of  HH    

  (.0013)    .0078***  (.0006)  

(.0012)    .0071***  (.0006)  

       Age2  of  Head  of  HH     -­‐.00007***   -­‐.00006***  

    (0.000005)   (0.000005)          

Head  of  HH  Married     .0152***   .0161***       (.0057)   (.0053)          

Head  of  HH  Female     -­‐.02507***   -­‐.0157***      

Head  of  HH  Education  

  (.0043)    -­‐.0066***  

(.0041)    -­‐.0041***  

   

Year  2014      

Constant      

PSU  Fixed  Effects    

Month  fixed  Effects    

R-­‐Squared    

         .0211***      (.0031)    No    No    0.009  

(.0004)    .0066  (.0053)    -­‐.0989***  (.0204)    No    No    0.0488  

(.00037)    .0124  (.0051)    -­‐.0129  (.0306)    Yes    Yes    0.2644    

N   12160   12160   12160  Table  4:  Standard  errors  are  reported  in  parentheses.  *  p  <  0.10,  **  p  <  0.05,  ***  p  <  0.01.  No  stars  implies  that  the  coefficient  is  insignificant.      As mentioned previously in the data section, one problematic issue is the possibility of

underreporting the true amount of remittances received, by the households who actually receive

high amount of remittances. In reality, it given it is not possible to investigate whether the

households truly report the amount of remittances that they receive given that while

interviewing them, one is not able to control whether they are telling the truth or not. Therefore,

as an attempt to overcome this problem, I extend the previous analysis by restricting the sample.

The sample restriction is done by excluding the highest remittance-receiver households, given

  26  

that they are more likely to underreport the high amount of remittances that they may receive.

The new sample now consists in 11499 observations, and the maximum amount of remittances

received is 9900000 old leks in year, equivalent to 7333,3 euros per year.

 Table  5.  Correlation  between  remittances  and  unwillingness  to  work  with  restricted  sample    

  (1)   (2)   (3)   (4)  Remittances  

   

Head  Age      

Head  Age2      

Head  Married  

.00867***  (.00055)  

.00828***  (.00055)    .00125***  (.00028)    -­‐.000011***  (2.56e-­‐06)      -­‐.005138**    (.0025936)  

.00842***  (.00056)    .001269***  (.000289)    -­‐.0000113  ***  (2.56e-­‐06)      -­‐.00509**  (.0025931)  

.00852***  (.0012)    .00119  ***  (.00031)    -­‐.0000103  ***  (2.64e-­‐06)      -­‐.0042789  (.0040261)  

                 Head  Female     -­‐.00178   -­‐.00172   -­‐.00161  

    (.00194)    

(.00194)   (.00242)  

Head  School        -­‐.00026    (.00016)  

-­‐.00026    (.00016)  

.000046  (.00020)  

         HH  Number  

   

Year  2014      

                       Constant  

             .0106***  (.00055)  

.00068  **  (.00034)    -­‐.01035***  (.00239)      -­‐.00448    (.00916)  

.00065  *  (.00034)    -­‐.010358  ***  (.00239)      -­‐.00320    (.0093)  

.000483  (.00037)    -­‐.00796**  (.00381)      .01998*  (.01181)  

         Month  Fixed  

Effects    

PSU  Fixed  Effects  

 No    No  

 No    No  

 Yes    No  

 Yes    Yes  

         N  

R-­‐Squared  F-­‐Statistics  

11499  0.0208  244.79  

11499  0.0260  38.33  

11499  0.0283  17.61  

11499  0.0845  

 Table  5:  Standard  errors  are  reported  in  parentheses.  *  p  <  0.10,  **  p  <  0.05,  ***  p  <  0.01.  No  stars  implies  that  the  coefficient  is  insignificant.  Estimations  with  restricted  sample,  if  yearly  remittances  are  equal  or  less  than  9900000  old  leks.    

  27  

The regression is based on the same equation as previously. In Table 5., column (1) shows the

results of the simple regression, column (2) shows the results of the multiple regression with

the additional control variables, column (3) shows the results of the multiple regression with

month dummies included, and column (4) shows the results of the multiple regression with

both month dummies and primary sampling unit dummies included. As it can be inferred by

the coefficients shown in the table, the results for remittances are very similar to the previous

results presented in Table 3. On the other hand, there are some changes that can be noticed on

the coefficients of the other explanatory variables: in this case, in the specifications presented

in column (2) and (3) the coefficient of schooling has become insignificant, whereas the

coefficient of head married has become significant. The negative sign implies that being

married is associated with a decrease in the unwillingness to work. This coefficient becomes

insignificant when adding primary sampling unit dummies. The main finding of the analysis

with the restricted sample, is that the coefficient of remittances continues to remain robust in

terms of statistical significance, sign, and magnitude, regardless the sample restriction.  7. Instrumental Variables approach 7.1. The instruments

In this subsection there will be presented the instruments used and the way they are constructed.

As an attempt towards finding a causal effect of remittances in the unwillingness to work, I use

instruments for the main explanatory variables, remittances, in order to overcome endogeneity

problems. I use separately three different instruments in the two-stage least-squares regression.

The first instrument is the exchange rate. The reason why I use this instrument is that the

exchange rate is correlated with the amount of remittances. An increase in the amount of

remittances affects the exchange rate by appreciating the domestic currency Lek towards the

foreign currencies e.g: Euro. The reason behind this is that the monetary inflows in Albania

that come from abroad are in foreign currency (mainly Euro, because the main share of

emigrants is in the Eurozone countries). This leads to an increase in the money supply in the

foreign currency, while the supply of the domestic currency remains the same. As the amount

of foreign currency in the Albanian money market increases relative to the amount of the

domestic currency, the domestic currency appreciates relative to the foreign currency. This is

an implication that there is a relationship between the amount of remittances inflow in the

country and the exchange rate. Moreover, the exchange rate is not related to the unwillingness

  28  

to work through any other channel except remittances. Thus, in a theoretical viewpoint, the

exchange rate is expected to be a good instrument for solving the endogeneity problems in the

regression of the unwillingness to work on remittances. The construction of the exchange rate

instrument is done by matching each month of the data with the monthly exchange rate of 2009

and 2014 respectively. The data on monthly exchange rate are obtained from the website of the

Bank of Albania (Bank of Albania, 2017).

The second instrument that I use is the distance from the land borders and the sea ports with

each Prefecture. The distance instruments easily satisfy the exogeneity condition: distance from

the land borders and sea ports are exogenously given and do not directly affect the dependent

variable. On the other hand, distance from the land borders and sea ports is positively correlated

with the migration patterns, and consequently with the amount of remittances that the

households of the migrants receive from them. Intuitively, the propensity to migrate is higher

in the areas that are close or allow for an easier access in the neighbouring countries. First of

all, people who live nearby the neighbouring countries can migrate easier in these countries,

even in the short-run or find a temporary job there. The proximity of their settlements with the

border facilitates the migration of these people and increases their incentives to migrate,

because they are more related to the neighbouring countries, not only through economic and

trade activities, but also through cultural exchanges. Also, the fact that many of family

members, relatives, or friends of these people might have migrated earlier, creates even more

incentives for these people to migrate due to their network and information regarding the

respective foreign country.

I construct the distance instrument by assigning the average distance of each Prefecture from

the border to the respective observations. The same procedure is followed for constructing the

sea distance instrument. Then, the total distance instrument, is constructed as a sum of both

land distance and sea distance instruments. I use the total distance instrument for the

estimations because people have massively migrated and continue to migrate using both routes.

Finally, I construct an interaction instrument, which is the interaction of the exchange rate

instrument and total distance instrument. The reason behind constructing this instrument is the

fact that both distance and exchange rate are correlated with remittances, and the only way that

these variables affect the unwillingness to work is through remittances. Therefore, besides

using each variable separately as an instrument, it would be interesting to observe the

interaction of both instruments as a new instrument.

  29  

7.2. First stage results

In this subsection there will be presented the first stage results. In Table 4. there are presented

the results of the first stage regressions for each instrument. In these regressions the dependent

variable is the standardized amount of remittances and the independent variables are the

respective instruments for each regression. There are two main conditions for the first stage to

be valid: the coefficient of the instrument has to be statistically significant, and the F-statistic

should be higher than 10 (Bosker, 2017). In column 1 there is presented the first stage of using

total distance as an instrument. The correlation of the total distance and the amount of

remittances is positive and significant. The F-statistic is 66.15, which is higher than 10. This

indicate that based on the first stage results only, in a statistical perspective, total distance

appears to be good instrument for remittances. Column 2, shows the results of the regression

of standardized amount of remittances on the exchange rate. The coefficient of the exchange

rate is negative and statistically significant at 1% level, and the F-statistic is much higher than

the rule of thumb 10, meaning that this is also a valid instrument from a statistical point of

view.

Table  6:  First  stage  results.  Dependent  variable:  Standardized  amount  of  remittances  

  (1)   (2)   (3)  Total  distance  

   

Exchange  rate      

Interaction      

.0014***  (.00017)  

         -­‐.02445***  (.00198)  

           .0000082***  (0.0000012)    

Constant    -­‐.2456***  (.0315)  

3.3356***  (.2713)  

-­‐.1970***  (.0314)  

       F-­‐statistic  

                                       Number              

66.15    12160      

151.27    12160      

42.91    12160      

Table  6:  Standard  errors  are  reported  in  parentheses.  *  p  <  0.10,  **  p  <  0.05,  ***  p  <  0.01.  No  stars  implies  that  the  coefficient  is  insignificant.    

In column 3, there are presented the results of the regression of remittances on the interaction

between total distance and total exchange rate. The coefficient is positive and significant at 1%

  30  

level, although the magnitude is very low compared to the magnitudes of the coefficients for

distance and exchange rate separately. Even in this case the F-statistic is sufficiently high,

implying that this instrument is also acceptable considering the statistical indicators.

7.3. Reduced form estimates

In this subsection there will be presented the results of the reduced form estimates. The reduced

form is basically the regression of the endogenous variable on the other control variables that

are included in the regression. In this case, remittances will be the dependent variable, and the

other control variables will be included in the regression as control variables. The reduced form

equation is used to test for partial correlation. While including the instrument, and the other

control variables in the regression, one can observe the correlation between the instrument and

the endogenous variable, while controlling for the effect that the other control variables may

have in this correlation (Wooldridge, Introductory Econometrics, A Modern Approach, 2009).

Table 7., presents the results of the reduced form estimations. In column 1 there are presented

the results of the reduced form with the interaction variable used as the main control variable.

The coefficient of this variable indicates that the partial correlation between the interaction

variable and remittances is positive and significant at 1% level. The R-squared indicates that

more than 13% of remittances is explained by the variables included in this regression. The

other variables that are statistically significant at 1 % level in this regression are: gender,

number of members in a household, and the year dummy. In contrast, age of the head of the

household, its quadratic term, and education of the head of the household are insignificant.

In column 2, there are shown the results of the reduced form with exchange rate used as the

main explanatory variable. In contrast with the results of column 1, the coefficient of exchange

rate is positive, but statistically insignificant. This implies that despite the fact that there is a

correlation between exchange rate and remittances, which was demonstrated in the first stage

estimations, while accounting for the correlation between the other control variables and

remittances, the relationship between exchange rate and remittances is no longer significant.

This is an indication that the exchange rate instrument is a weak instrument, and therefore may

be problematic for the final outcome, if used as an instrument.

 

  31  

Table  7:  Reduced  form  results,  three  different  instruments  included  separately       (1-­‐Interaction)   (2-­‐Exchange  rate)   (3-­‐Total  distance)  

Instrument    

         Age  of  Head  of  HH  

 

.000027***  (.0000055)    .0048  (.0036)  

.0035  (.0046)    .0054  (.0036)      

.0037***  (.0007)    .0047  (.0035)  

Age2  of  Head  of  HH    

-­‐.00003  (.00003)  

-­‐.000037  (.00003)  

-­‐.00003  (.00003)  

   Head  of  HH  married  

 -­‐.0194  

 -­‐.00865  

 -­‐.0204  

  (.0562)   (.0555)   (.0564)    

Head  of  HH  Female    

 .1013***  (.0360)  

 .1045***  (.0369)  

 .1011***  (.0360)  

 Head  of  HH  Education      

 -­‐.0031  (.0027)  

 -­‐.0047*  (.0027)  

 -­‐.0031  (.0027)  

 Number  of  HH  

members    

Year  2014  

 .10411***  (.0225)    -­‐.3858***  (.0755)  

 .1082***  (.0228)    -­‐.3227***  (.0668)  

     .104***  (.0225)    -­‐.3485***  (.0709)  

 Constant  

   

PSU  fixed  effects    

R-­‐squared      

     -­‐.5544***  (.1947)    Yes    0.1345  

 -­‐.4194  (.6226)    Yes    0.1263  

 -­‐.6182***  (.2045)    Yes    0.1346  

N    

12160    

12160    

12160    

Table  7:  Standard  errors  are  reported  in  parentheses.  *  p  <  0.10,  **  p  <  0.05,  ***  p  <  0.01.  No  stars  implies  that  the  coefficient  is  insignificant.    

In column 3, there are presented the results of the reduced form estimates with total distance

used as the main explanatory variable. The coefficient of total distance is positive and

approximately 137 times higher in magnitude compared to the interaction coefficient. This

coefficient is significant at 1% level, implying that this is not the case of a weak instrument.

The coefficients of the other control variables are very similar in sign, magnitude, and statistical

significance, with the coefficients of the reduced form equation presented in column 1. The

same similarity can be observed in the magnitude of the R-squared. Overall, the estimations of

the reduced form with the three types of specifications, infer that the partial correlation is

  32  

present between remittances and interaction term, and remittances and total distance term,

whereas it is not significant in the case of exchange rate, meaning that the exchange rate should

be considered as a weak instrument. In section 8, there will be presented further arguments

regarding the problems that are related to the exchange rate instrument.

7.4. Second stage results

In this subsection there are presented the results of the estimations of the second stage, using

IV approach. Table 7., presents the results of the second stage of regression, using each

instrument separately, and their interaction. The estimations presented in this table show the

results of the second stage of the IV regression without including neither PSU fixed effects,

nor Month fixed effects. Column 1 in Table 8., shows the results of the second stage of the IV

regression while using the interaction between exchange rate and distance as an instrument.

The coefficient of remittances is positive and significant at 5% level. The magnitude of this

coefficient is 2.7 times higher compared to the previous results of pooled OLS with and without

controlling for Month and PSU fixed effects. Also, the standard errors of this coefficient are

almost 18 higher compared to the OLS coefficient. This is associated with a wide range of the

confidence interval at 95% confidence level, whose values are between 0.0028 and 0.045.

As argued in (Wooldridge, Introductory Econometrics, A Modern Approach, 2009), this is not

an unexpected outcome while trying to obtain a good estimator by assuming the endogeneity

of remittances. The coefficients of age of the head of household, and the quadratic term of age,

are very similar to the respective coefficients in the pooled OLS regression, without including

month and PSU dummies. The coefficients that indicate marital status, gender, number of

school years, and number of household members, are insignificant. This is in contrast with the

previous results, where the coefficient on education was significant at 5% level, and the

coefficient on the number of family members was significant at 10% level. Finally, the

coefficient of year 2014 is negative and significant at 5% level, indicating that year 2014 is

negatively correlated with the unwillingness to work. This is in line with the expectations, and

as will be explained in further sections, year 2009 is associated with higher remittance inflows,

exchange rate appreciation, and increase of the unwillingness to work.

  33  

Table  8:  Second  stage  results  using  each  instrument  separately

  (1-­‐Interaction  Instrument)  

(2-­‐Exchange  rate  instrument)  

(3-­‐Distance  instrument)  

Remittances      

Age  of  Head  of  HH      

Age2  of  Head  of  HH    

.0268**  (.0109)    .0014***    (.0003)    -­‐.000013***    (.000003)  

.3173  (.7866)    .00039    (.0032)    -­‐0.000005  (.000024)  

.02504**  (.0097)    .0014***  (.00032)    -­‐.000013***  (0.0000028)  

             Head  of  HH  married   -­‐.0041   -­‐.0106   -­‐.00407  

  (.0028)   (.0218)   (.00282)          

Head  of  HH  Education       -­‐.00028   .0019   -­‐.00029     (.00019)   (.0061)   (.00019)          

Head  of  HH  Female      

Year  2014      

Number  of  HH  members  

   

                                         Constant      

PSU  fixed  effects  

-­‐.0026    (.0024)      -­‐.0088**  (.0034)    -­‐.0013  (.0012)      -­‐.0009  (.0102)    No  

-­‐.0348  (.0877)    .0577  (.1807)    -­‐.0340  (.0886)      .0426  (.1267)    No  

 -­‐.0024  (.0024)    -­‐.0092***  (.0034)        -­‐.0011  (.00116)          -­‐.00116  (.01012)    No  

       N    

12160   12160   12160  

       Table8:  Standard  errors  are  reported  in  parentheses.  *  p  <  0.10,  **  p  <  0.05,  ***  p  <  0.01.  No  stars  implies  that  the  coefficient  is  insignificant.      

In column 2 there are presented the results of the second stage of the IV regression, with the

exchange rate instrument. All the coefficients in this regression are statistically insignificant,

meaning that using the exchange rate instrument separately, makes the regression estimates

meaningless and leads to inefficient results. In the next section, I will elaborate further

regarding the problems that this instrument is associated with.

  34  

Column 3 shows the results of the second stage of IV regression, while using total distance as

an instrument. The coefficient of remittances is significant at 5% level, and its magnitude is

very similar to the respective coefficient in column 1. The magnitude has been slightly

decreased, as well as the standard errors. Also, the other coefficients of the control variables,

are very similar with the coefficients of the regression with the interaction term as an

instrument, in both terms of magnitude and statistical significance. The only difference is in

the coefficient of year 2014, which is now significant at 1% level, and slightly higher in

magnitude.

   In Table 9., there are shown the results of the second stage results while including Primary

Sampling Unit dummies, as an attempt to capture PSU fixed effects. In contrast with the results

presented in Table 7., where PSU dummies were not included, the coefficients of remittances

are insignificant in all the regression specifications, regardless the instrument used. This

suggests that the inclusion of PSU fixed effects, is associated with a decrease in the magnitude

of the coefficient of remittances, which in turn decreases the z-statistic, and leads to statistically

insignificant estimates.

 

8. Problems with the exchange rate instrument

The results in the previous subsection demonstrated that when using exchange rate as

instrument, all the coefficients of the control variables were insignificant. On the other hand,

when using only distance as an instrument, or the interaction between distance and the

exchange rate, the coefficient of remittances was positive and statistically significant, as

expected. The problems with the exchange rate variable can be observed firstly in the reduced

form results. The insignificant of the coefficient of exchange rate in the reduced form, implies

that the partial correlation between exchange rate and remittances is no longer present after the

inclusion of the other control variables in the regression. This indicates that the exchange rate

variable is a weak instrument, and this is problematic for the final outcome.

 

 

 

 

 

  35  

Table  9:  Second  stage  results  using  each  instrument  separately

  (1-­‐Interaction  Instrument)  

(2-­‐Exchange  rate  instrument)  

(3-­‐Distance  instrument)  

Remittances      

Age  of  Head  of  HH      

Age2  of  Head  of  HH    

.0079  (.0062)    .0014***    (.00031)    -­‐.000012***    (.000003)  

.1257  (.1398)    .00075    (.0009)    -­‐0.0000075  (.00000758)  

   .0064  (.0062)    .0014***  (.0003)    -­‐.000012***  (0.0000028)  

             Head  of  HH  married   -­‐.0029   -­‐.0019   -­‐.0029  

  (.0027)   (.0056)   (.00282)          

Head  of  HH  Education       -­‐.00008   .00048   -­‐.00008     (  .00019)   (.00076)   (.00019)          

Head  of  HH  Female      

Year  2014      

Number  of  HH  members  

   

                                         Constant      

PSU  fixed  effects  

-­‐.00062    (.0022)      -­‐.0105***  (.0032)    .00051  (.00077)      .0181  (.0156)    Yes  

-­‐.0129  (.0152)    .0243  (.0417)    -­‐.0123  (.0152)      .0168  (.0313)    Yes  

 -­‐.00047  (.0022)        -­‐.0109***  (.0032)        .00066  (.00077)          .0181  (.0156)    Yes  

       N    12160   12160   12160  

       Table9:  Standard  errors  are  reported  in  parentheses.  *  p  <  0.10,  **  p  <  0.05,  ***  p  <  0.01.  No  stars  implies  that  the  coefficient  is  insignificant.  Standard  errors  are  clustered  by  PSU.    

There are possible explanations that can be provided regarding the problems that associate the

use of the exchange rate as an instrument. In contrast with the geographical distance, which is

constant over time and is correlated with migration and consequently with remittances,

exchange rate is volatile and it changes with the change of the amount of remittances. Several

empirical evidences demonstrate that an increase in remittance inflows leads to an appreciation

of the real exchange rate. (Lopez, Molina, & Bussolo, 2007) find that for several Latin

  36  

American countries the increase in the amount of remittances inflows in these countries, is

associated with an appreciation of the real exchange rate. The same pattern is confirmed by

(Lartey, Mandelman, & Acosta, 2008), who use a panel of 109 countries during the time period

1990-2003 in their analysis. Taking these findings into consideration, a similar association can

be found for the case of Albania. The total amount of remittance inflows from 2009 to 2014

has been decreased with almost 150 million euros in Albania, reaching the minimum in 2013.  

In the meantime, the exchange rate in 2014 was depreciated compared to 2009: the average

annual exchange rate in 2009 was 132.19 Lek/Euro, and it turned into 139.97 Lek/Euro in 2014

(Bank of Albania, 2017). At a first sight, the correlation between the increase in the remittances

inflows and exchange rate appreciation is evident even for the case of Albania for years 2009

and 2014. In addition, in order to provide further evidence about this correlation, I run two

simple OLS regressions where I regress the standardized amount of remittances on the

exchange rate, and the exchange rate on the standardized amount of remittances. I use the full

sample with 42426 observations for these estimations.

Table 10. presents the results of these regressions. When the dependent variable is the

standardized amount of remittances (column 1), the coefficient of the exchange rate is negative

and significant at 1% level. The same sign and significance is in the case when the dependent

variable is the exchange rate, and the control variable is the standardized amount of

remittances. The significance at 1% level in both regressions demonstrates that the relationship

between the amount of remittances and exchange rate is statistically significant, and the

negative sign of the coefficients implies that this correlation is in line with the previous

findings. When observing the sign of this correlation, one should keep in mind that the

exchange rate is expressed as: X(Lek/Euro), where X represents the amount of domestic

currency Lek, relative to one Euro. Therefore, the negative correlation coefficient found in the

data, and the previous findings in other remittance-receiver countries, indicate that an increase

in the amount of remittance inflows in the country, decreases the value of X. This suggests that

one Euro will be exchanged with a lower amount of the domestic currency, which is an

appreciation of the domestic exchange rate.

         

  37  

Table  10:  Correlation  between  the  standardized  amount  of  remittances  and  the  exchange  rate(OLS)    

  (1)   Dependent  variable:  remittances  

(2)   Dependent  variable:  Exchange  rate  

 Remittances    

                                                                                                                   Exchange  rate  

   

Constant      

N  

                             -­‐.0323***                        (.0011)                          4.4133***                        (.1464)                            42426                      

                         -­‐.6508***                          (.0215)                                    136.84***                            (.0215)                              42426  

     F-­‐statistic                        909.67                            909.67  

     R-­‐squared                        0.021                            0.021  

     Table  10:  Standard  errors  are  reported  in  parentheses.  *  p  <  0.10,  **  p  <  0.05,  ***  p  <  0.01.  No  stars  implies  that  the  coefficient  is  insignificant.      After clarifying the correlation between the amount of remittance inflows in the country and

the appreciation of exchange rate, it is more simple to provide an explanation regarding the

problems that the exchange rate instrument presented. First of all, there has been substantial

changes in the amount of remittances inflows from 2009 to 2014, which is associated with

exchange rate depreciation. The decline in the total amount of remittances, where the minimum

is reached in 2013, can partially be attributed to the Greek crisis. During the Greek crisis, a

considerable number of Albanian emigrants, who lived and worked in Greece, decided to return

in Albania and withdraw their savings from their Greek Bank accounts. The withdrawal process

was present starting from 2008, and is estimated to have continued even in the subsequent years

(ACIT, 2012). Taking into consideration the fact that the economic conditions were not

deteriorated only in Greece, but also in other Eurozone countries, such as Italy and Spain, and

that the largest share of Albanian emigrants is in Greece and Italy, the decrease in the amount

of remittances from 2009 to 2014, can be explained by these events. On the other hand, given

that from the beginning of the Greek crisis, many of Albanian migrants withdrew their savings

and sent them as remittances in Albania, can explain the fact that the amount of remittance

inflows declined gradually from the beginning of the crisis, and reached the minimum in 2013.

These considerable changes create volatility in the data of exchange rate and amount of

remittances of year 2009 relative to year 2014. Moreover, the share of the data from 2014 is

  38  

larger than the share of the data from 2009, which makes the final outcome to be driven more

from the 2014 results. Another problem with the use of the exchange rate as an instrument, is

the relationship that it has with the macro fundamentals, which may influence individual saving

decisions. Consequently, this may create another channel of correlation between the exchange

rate and unwillingness to work. The reason behind this is that unwillingness to work may be

correlated with savings, and savings are influenced by the interest rate, which is correlated with

the exchange rate.

9. Conclusions

This study presents an attempt towards finding a causal effect of remittances on the individuals’

unwillingness to work. The first estimations are conducted using Pooled OLS regression, given

that the data are not panel data, and therefore do not allow for using other models such as fixed

effects. The OLS results provide information regarding the correlation between the control

variables and the dependent variable, which indicate that there is a positive and significant

relationship between the remittances and unwillingness to work of the households. The

magnitude of the coefficient of remittances suggests that an increase in the standard deviation

of remittances with one unit, is associated with an increase with 0.009 units of the

unwillingness to work. The results are robust after the inclusion of month and PSU dummies,

which are used for capturing month and PSU fixed effects. Then, as an attempt to avoid omitted

variable bias and endogeneity problems, I estimate the econometric model using Instrumental

Variables with and without including PSU dummies. The results of the IV are consistent only

in the cases where total distance of the interaction term are used as instruments, and infer that

the IV coefficient is positive and higher than the OLS coefficient. On the other hand, the use

of exchange rate as an instrument is problematic. This is confirmed by the reduced form results

and the second stage results. These problems with the exchange rate can be attributed to the

high exchange rate volatility between years 2009 and 2014, and the differences in the

observation numbers in these years. Also, the IV results are not robust after the inclusion of

PSU dummies.

However, despite the problems and limitations of the analysis, the IV procedure while using

total distance, and interaction as an instrument, without including PSU dummies, indicates that

the amount of remittances positively affects the unwillingness to work, which is consistent with

the findings in the previous regressions.

  39  

The consistent result of the positive impact of remittances in the unwillingness to work for the

Albanian remittance-receiver individuals, indicates that for these households remittances play

the role of an additional source of income. Therefore, their demand for leisure increases and

the alternative options such as: taking care of the children, or maintaining the dwelling, become

more attractive than working, which explains the increase of the unwillingness to work.

The main finding can be explained also through the labour market environment. As shown in

the previous sections, Albania is a country with continuous levels of high unemployment. This

is an indicator of a problematic labour market, where labour supply is higher than the labour

demand, and consequently the wages are not high. Therefore, the incentives to work for the

remittance-receiver individuals are low, considering the low wages offered in the labour

market. The fact that remittances are associated with an increase in the unwillingness to work,

can be interpreted in two ways in terms of the consequences for the Albanian economy. The

first interpretation would be that the increase in the unwillingness to work is a negative

outcome, which can be translated into a decrease of the labour force of the country. On the

other hand, given the high unemployment rates in Albania, keeping a part of people out of the

labour force, and providing an additional source of income to them through remittances, can

be considered as an “opportunity” for them to not become unemployed. Instead of constantly

looking for a job and not finding it, which would be associated with lack of income and poverty

for these individuals, they can cover their expenditures through remittances, and thus not face

poverty and hard living conditions. However, in a long-term perspective the increase of the

unwillingness to work is not a good indicator for the economy. As a way to solve this problem

there can be designed policies that increase the incentives of the individuals for creating and

developing microenterprises. A better environment for microenterprises would create

incentives for remittance-receiver households to invest the amounts of remittances and

generate more income, instead of using them as an additional source of income in absence of

wages.

  40  

 

 

10.  Bibliography  Çela,  A.,  Kamberi,  G.,  &  Pici,  E.  (2015).  Albanian  Youth  2015  "Slow  Change,  Internet  

Dependency  and...EU  trust!".  Friederich  Ebert  Stiftung.  ACIT.  (2012,  August).  Study  on  the  Economic  Impact  of  the  Greek  Crisis  in  Albania.  Retrieved  

from  ACIT,  USAID:  https://www.usaid.gov/sites/default/files/documents/1863/USAID%20Study%20on%20Greek%20Crisis.pdf  

Acosta,  P.  A.,  Lartey,  E.  K.,  &  Mandelman,  F.  S.  (2007,  April).  Remittances  and  the  Dutch  Disease.  Retrieved  from  Federal  Reserve  Bank  of  Atlanta:  https://www.frbatlanta.org/-­‐/media/Documents/filelegacydocs/wp0708.pdf  

Airola,  J.  (2008).  Labor  Supply  in  Response  to  Remittance  Income:  The  Case  of  Mexico.  The  Journal  of  Developing  Areas,  41,  69-­‐78.  

Bajraba,  K.,  &  Bajraba,  J.  (2015,  September  10).  Migration  Information  Source.  Retrieved  from  Migration  Policy  Institute  :  http://www.migrationpolicy.org/article/embracing-­‐emigration-­‐migration-­‐development-­‐nexus-­‐albania  

Bank  of  Albania.  (2017).  Monetary  Policy,  Exchange  Rates  Archive.  Retrieved  from  Bank  of  Albania:  https://www.bankofalbania.org/web/exchange_rates_archive_2372_2.php?kc=0,2,3,5,0  

Bank  of  Albania.  (2017).  Statistics,  Time  Series  Database  of  the  Bank  of  Albania.  Retrieved  from  Bank  of  Albania:  https://www.bankofalbania.org/?evn=agregate_detaje&evb=agregate&cregtab_id=536&periudha_id=5&crd=0,3,4,0,0,22&ln=1  

Bosker,  M.  (2017).  Applied  Econometrics:  A  Crash  Course.  Empirics  of  Trade  and  Development  .  Rotterdam:  EUR-­‐ESE.  

Cambridge  Dictionary.  (2017).  transition  economy.  Retrieved  2017,  from  Cambridge  Dictionary:  http://dictionary.cambridge.org/us/dictionary/english/transition-­‐economy  

Carlo  Azzarri,  Calogero  Corletto.  (2009,  May).  Modeling  Migration  Dynamics  in  Albania,  A  Hazard  Function  Approach.  Retrieved  from  World  Bank:  http://documents.worldbank.org/curated/en/909861468006293697/pdf/WPS4945.pdf  

Deaton,  A.  S.  (2009,  January).  Instruments  of  Development:  Randomization  in  the  Tropics,  and  the  Search  for  the  Elusive  Keys  to  Economic  Development.  National  Bureau  of  Economic  Research,  12-­‐14.  

Dermendzhieva,  Z.  (2010).  Migration,  remittances,  and  labor  supply  in  Albania.  Retrieved  from  The  World  Bank:  http://wbox.cerge-­‐ei.cz/pdf/gdn/RRCVIII_60_paper_01.pdf  

Gedeshi,  I.  (2002).  Role  of  Remittances  from  Albanian  Emigrants  and  Their  Influence  in  the  Country's  Economy.  Eastern  European  Economics,  40(5),  49-­‐72.  

Giuliano,  P.,  &  Ruiz-­‐Arranz,  M.  (2005).  Remittances,  Financial  Development,  and  Growth.  IMF.  

  41  

Gupta,  S.,  Pattillo,  C.,  &  Wagh,  S.  (2007).  Impact  of  Remittances  on  Poverty  and  Financial  Development  in  Sub-­‐Saharan  Africa.  IMF.  

INSTAT.  (2017).  Labour  Market.  Retrieved  from  INSTAT:  http://www.instat.gov.al/al/themes/tregu-­‐i-­‐pun%C3%ABs.aspx  

INSTAT.  (2017).  Labour  Market,  Publications.  Retrieved  from  INSTAT:  http://www.instat.gov.al/al/themes/tregu-­‐i-­‐punës/publications/books.aspx  

Jarvis,  C.  (1999).  The  Rise  and  Fall  of  the  Pyramid  Schemes  in  Albania.  Retrieved  from  International  Monetary  Fund:  https://poseidon01.ssrn.com/delivery.php?ID=024094113006078093076095082087004102067074018037042059108036062109006100020035023036026086017008115018015102019087008084106119125002031021088070004088119003085110122002111105022014070115118119092&EXT=pdf  

Kim,  N.  (2007,  February).  The  Impact  of  Remittances  on  Labor  Supply:  The  case  of  Jamaica.  Retrieved  from  World  Bank:  https://openknowledge.worldbank.org/handle/10986/7152?show=full  

Konica,  N.,  &  Filer,  R.  K.  (2009).  Albanian  Emigration:  Causes  and  Consequences.  South-­‐Eastern  Europe  Journal  of  Economics  1,  75-­‐98.  

Lartey,  E.  K.,  Mandelman,  F.  S.,  &  Acosta,  P.  A.  (2008,  March).  Remittances,  Exchange  Rate  Regimes,  and  the  Dutch  Disease:  A  Panel  Data  Analysis.  Retrieved  from  Federal  Reserve  Bank  of  Atlanta:  https://www.frbatlanta.org/-­‐/media/Documents/research/publications/wp/2008/wp0812.pdf?la=en  

Lopez,  H.,  Molina,  L.,  &  Bussolo,  M.  (2007,  April).  Remittances  and  the  real  exchange  rate.  Retrieved  from  World  Bank  Policy  Research  Working  Paper  4213,  April  2007:  https://poseidon01.ssrn.com/delivery.php?ID=679064026081101078127098116020113078035010031014027056097013062086030037000020022000123003074006088029018009055007111127105110116120027017111127077028124086096001004080092020090014095066022&EXT=pdf  

Meyer,  D.,  &  Shera,  A.  (2016,  June).  The  impact  of  remittances  on  economic  growth:  An  econometric  model.  Retrieved  from  EconomiA:  http://dx.doi.org/10.1016/j.econ.2016.06.001  

Rodriguez,  E.  R.,  &  Tiongson,  E.  R.  (2001).  Temporary  Migration  Overseas  and  Household  Labor  Supply:  Evidence  from  Urban  Philippines.  The  International  Migration  Review,  709-­‐725.  

United  Nations.  (2014).  Country  Classification.  Retrieved  from  United  Nations:  http://www.un.org/en/development/desa/policy/wesp/wesp_current/2014wesp_country_classification.pdf  

United  Nations.  (2016).  International  Migrant  Stock  2015.  Retrieved  from  United  Nations,  Department  of  Economic  and  Social  Affairs:  http://www.un.org/en/development/desa/population/migration/data/estimates2/estimates15.shtml  

Uruçi,  E.,  &  Gedeshi,  I.  (2004).  Role  of  Remittances  in  the  Social  Economic  Development  of  the  Country  .  Bank  of  Albania.  

Wooldridge,  J.  M.  (2009).  Introductory  Econometrics,  A  Modern  Approach.  In  J.  M.  Wooldridge,  Introductory  Econometrics,  A  Modern  Approach  (Vol.  4,  p.  527).  South-­‐Western  CENGAGE  Learning.  

Wooldridge,  J.  M.  (2013).  Introductory  Econometrics,  A  Modern  Approach.  Michigan  State  University.  

  42  

World  Bank.  (2016).  Data  Bank,  World  Development  Indicators.  Retrieved  from  The  World  Bank:  http://databank.worldbank.org/data/reports.aspx?source=world-­‐development-­‐indicators  

World  Bank.  (2017).  Understanding  Poverty/Topics/Migration/Overview.  Retrieved  from  World  Bank:  http://www.worldbank.org/en/topic/migrationremittancesdiasporaissues/overview  

                                                                                         

  43  

11.  Appendix    Table  1:  Hausman  test  for  endogeneity  of  remittances  

  (b)  all  cats  

(B)    

(b-­‐B)  difference   Standard  errors  

Age  of  Head  of  HH  

Age2  of  Head  of  HH  

.0036    -­‐.00003  

.0015    -­‐.000013  

.0021    -­‐.000013  

.0046    .000041  

Head  of  HH  Married  

.0223   -­‐.0037      .02602   .04103  

Head  of  HH  Female  

.1108   -­‐.00063   .1114   .0312  

Education  of  Head  of  HH  

-­‐.0076   -­‐.00042   -­‐.0072   .0026  

Number  of  HH  members  

.1126   .00068   .1119   .0054  

Year  2014   -­‐.2292   -­‐.0129   -­‐.2162   .0380            

 H0:  Difference  in  coefficients  is  not  systematic    Prob>Chi2=0.0000  The null hypothesis is rejected, meaning that the differences in coefficients are statistically

significant, and remittances is endogenous.